DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
MMGist: A Comprehensive Multimodal Benchmark for 20272h◆Visualizing "We the People": Bridging the Perception Gap through Pluralistic Data Storytelling2h◆Small edits, large models: How Wikipedia advocacy shapes LLM values2h◆Noise-Aware Boundary-Enhanced Generative Learning for Ultrasound Speckle Reduction2h◆Wan-Streamer v0.1: End-to-end Real-time Interactive Foundation Models2h◆MiniOpt: Reasoning to Model and Solve General Optimization Problems with Limited Resources2h◆HierBias: Context-Conditioned Hierarchical Media Bias Detection with Multi-Task Type Classification2h◆Where Larger Models Excel: The Primacy of Constraint-Guided Reasoning2h◆Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM2h◆Phonetic and semantic analyses of spoken corpora of Beijing and Taiwan Mandarin indicate that the neutral tone is a lexical tone2h◆ProfileFoundry: A Synthetic Person-Object Substrate for Privacy, Memory, and Tool-Use Evaluation in LLM Agent2h◆AnySimLite: A Lightweight Few-Shot Similarity Encoder for On-Device Speech-Adjacent Classification2h◆Extracting Problem and Method Sentence from Scientific Papers: A Context-enhanced Transformer Using Formulaic Expression Desensitization2h◆Utilizing Cognitive Signals Generated during Human Reading to Enhance Keyphrase Extraction from Microblogs2h◆Comparing BERT Sentence-Pair Classification and Few-Shot LLM Prompting for Detecting Threat and Solution Framing in German Climate News2h◆Nemotron-TwoTower: Diffusion Language Modeling with Pretrained Autoregressive Context2h◆Assessing Post-Reform Changes in Risk Disclosure Quality with a Multidimensional Text Analysis Approach2h◆Erase-then-Delta Attention: Decoupling Erase and Write Addresses in Delta-Rule Linear Attention2h◆Zero-shot Tweet-Level Stance Detection Enhanced by External Knowledge and Reflective Chain-of-Thought Reasoning2h◆Closing the Quality Gap in Low-Resource Text-to-Speech: LoRA Fine-Tuning of VoxCPM2 for Khmer and Korean2h◆MMGist: A Comprehensive Multimodal Benchmark for 20272h◆Visualizing "We the People": Bridging the Perception Gap through Pluralistic Data Storytelling2h◆Small edits, large models: How Wikipedia advocacy shapes LLM values2h◆Noise-Aware Boundary-Enhanced Generative Learning for Ultrasound Speckle Reduction2h◆Wan-Streamer v0.1: End-to-end Real-time Interactive Foundation Models2h◆MiniOpt: Reasoning to Model and Solve General Optimization Problems with Limited Resources2h◆HierBias: Context-Conditioned Hierarchical Media Bias Detection with Multi-Task Type Classification2h◆Where Larger Models Excel: The Primacy of Constraint-Guided Reasoning2h◆Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM2h◆Phonetic and semantic analyses of spoken corpora of Beijing and Taiwan Mandarin indicate that the neutral tone is a lexical tone2h◆ProfileFoundry: A Synthetic Person-Object Substrate for Privacy, Memory, and Tool-Use Evaluation in LLM Agent2h◆AnySimLite: A Lightweight Few-Shot Similarity Encoder for On-Device Speech-Adjacent Classification2h◆Extracting Problem and Method Sentence from Scientific Papers: A Context-enhanced Transformer Using Formulaic Expression Desensitization2h◆Utilizing Cognitive Signals Generated during Human Reading to Enhance Keyphrase Extraction from Microblogs2h◆Comparing BERT Sentence-Pair Classification and Few-Shot LLM Prompting for Detecting Threat and Solution Framing in German Climate News2h◆Nemotron-TwoTower: Diffusion Language Modeling with Pretrained Autoregressive Context2h◆Assessing Post-Reform Changes in Risk Disclosure Quality with a Multidimensional Text Analysis Approach2h◆Erase-then-Delta Attention: Decoupling Erase and Write Addresses in Delta-Rule Linear Attention2h◆Zero-shot Tweet-Level Stance Detection Enhanced by External Knowledge and Reflective Chain-of-Thought Reasoning2h◆Closing the Quality Gap in Low-Resource Text-to-Speech: LoRA Fine-Tuning of VoxCPM2 for Khmer and Korean2h◆
Databubble·

Newsletter

The Bubble Brief
Weekly · Tue 13:00 UTC

// a weekly bloomberg-style brief of the AI model market

THE BUBBLE
BRIEF.

Five minutes. Tuesday 13:00 UTC. The week's top movers, one story explained, a price watch, every new model tracked. Original data analysis only — no source paraphrasing, no AI fluff.

Subscribe
Step 01 / 02

Free forever. One click to unsubscribe at any time.

// no spam · unsubscribe one-click · free forever

What you'll get
5 sections
  • 01
    Top movers

    The 5 models with the largest 7-day shift — downloads, arena ELO, or benchmarks. Ranked, with stats.

  • 02
    Cooling off

    Models that lost steam. Surfaces fading momentum before it shows up on the leaderboards.

  • 03
    Story of the week

    One signal, explained. Why it matters, what it changes, who it touches. ~200 words, no fluff.

  • 04
    Price watch

    Per-million-token pricing changes across hosted APIs. Diffs from last week, with absolute numbers.

  • 05
    New models tracked

    Every new model added to the universe this week, with HF id and provider.

Sample issue
WK 18 · 04 May 2026
=========================================================
THE BUBBLE BRIEF · WK 18 · 04 May 2026
=========================================================

[01] TOP 5 MOVERS  · 7d
   01  deepseek-v4-pro            ▲  +28.3%   DL  1.2M
   02  meta-llama/llama-4-405b    ▲  +14.7%   DL    684K
   03  qwen/qwen-3-72b-instruct   ▲  +11.2%   DL    503K
   04  mistralai/mixtral-8x32b    ▲   +9.4%   DL    441K
   05  google/gemma-3-27b         ▲   +6.8%   DL    312K

[02] COOLING OFF
   01  meta-llama/llama-3-70b     ▼   -8.1%   DL    229K
   02  cohere/command-r-plus      ▼   -5.4%   DL     91K
   03  microsoft/phi-3-medium     ▼   -3.9%   DL    142K

[03] STORY OF THE WEEK
   DeepSeek V4 lands. 405B MoE, 32K active, MIT-license.
   Arena debut at 1,348 ELO — 4 points behind GPT-5,
   2 points ahead of Claude 4.5. Pricing: $0.27 / Mtok in,
   $1.10 / Mtok out. The aggressive open-weight pricing
   is the headline; the architecture (sparse MoE with
   expert-routing on long-context) is the real signal.

[04] PRICE WATCH  · per-million tokens
                                 IN $        OUT $       Δ 7d
   01  openai / gpt-5-mini       0.45        1.80         —
   02  anthropic / claude-4.5    3.00       15.00        −10%
   03  deepseek / deepseek-v4    0.27        1.10        new
   04  google / gemini-2.5       0.50        2.00         —
   05  mistral / mistral-large   1.20        4.80         −8%

[05] NEW MODELS TRACKED  · 14 added this week
   - deepseek-ai/DeepSeek-V4-Pro
   - alibaba/Qwen3-Coder-32B
   - cohere/aya-32b-multilingual
   - 11 more — see the bubble chart →

=========================================================
that's it for this week. tuesday 13:00 utc, see you then.
=========================================================

// sample only · numbers above are illustrative

The fine print
  • // no spam — one email per week, on Tuesday
  • // unsubscribe with one click in every issue
  • // free forever — no paywall on the brief
  • // ad-free in your inbox — promise
  • // founded 2026 · independent project
  • // data sourced from the Databubble universe
← back to news·how the data is computed →
HomeModelsNews